Documentation

Creating Agents

Step-by-step guide to creating your first AI agent in Kindship.

Creating an agent in Kindship is simple — the system guides you through each step. This guide covers everything you need to know to create an effective agent.

Before You Start

Think about what you want your agent to help with:

  • What's your main goal? — Having a clear purpose helps your agent be more effective
  • What kind of tasks? — Planning, research, creative work, project management, etc.
  • What's your working style? — Do you prefer detailed responses or quick summaries?

You don't need to have everything figured out — your agent will help you refine these as you go.

Creating Your Agent

Step 1: Start the Creation Process

From your dashboard, click the Create Agent button. You might also see "Create My Agent" or a similar option depending on your account type.

Your agent will be initialized with its own workspace. This happens automatically — no technical configuration required.

New-agent setup now focuses on the choices that matter most upfront. More specialized settings, such as coding credentials, model selection, communication oversight, and public channels, can be reviewed after the workspace exists.

Step 2: Define Your Agent's Purpose

Once created, you'll be in the chat interface with your new agent. Start by telling it what you want to accomplish:

Example: "I want you to help me launch a new product. I need help with market research, planning the launch timeline, and creating marketing content."

Your agent will ask clarifying questions to understand your needs better:

  • What kind of product?
  • What's your timeline?
  • Who's your target audience?
  • Do you have existing resources?

Answer these naturally — there's no special format required.

Step 3: Set Up the Prime Directive

As you discuss your goals, your agent will help you craft a Prime Directive — a clear statement of purpose stored in PRIME_DIRECTIVE.md on the agent's container. This file guides all future interactions.

A good Prime Directive includes:

  • Who you are (or your team/company)
  • What you're trying to achieve
  • Core values or principles
  • Any important constraints

Example Prime Directive: "Help [Name] successfully launch their educational app for language learners. Focus on reaching adult learners aged 25-45 who want practical, time-efficient learning. Prioritize data-driven decisions and ethical marketing practices."

Your agent references this directive when helping you make decisions.

Step 4: Establish Initial Goals

Work with your agent to set up your initial projects and processes. This creates structure for your collaboration:

Projects — Specific initiatives

  • Market research
  • Beta testing program
  • Launch campaign
  • Customer support setup

Processes — Recurring workflows

  • Daily progress check-in
  • Weekly metrics review

Your agent will help you break these down further as you work together.

Customizing Ethical Guardrails

During agent creation, you can configure which ethical guardrails apply to your agent. Guardrails are runtime constraints that guide how your agent approaches decisions, interactions, and recommendations.

Choosing Your Configuration

You have two options:

  • Standard (Recommended) — Includes all mandatory, recommended, and Kindship guardrails. This is the best choice for most users.
  • Customize — Fine-tune which guardrails are active for your specific needs.

Guardrail Categories

Mandatory — Always active for safety and compliance. These cannot be disabled and include identity transparency, human oversight for high-stakes decisions, and protection of minors.

Recommended — Enabled by default but can be customized. Covers epistemic humility, anti-stereotyping, cultural inclusivity, and privacy preservation.

Kindship — Principles for authentic AI relationships: identity through change, no fake feelings, knowing when to ask, and dignity in endings.

Voluntary — Opt-in guardrails for specialized needs, including indigenous data sovereignty (OCAP), care ethics, virtue ethics, and philosophical frameworks like Kant's Categorical Imperative.

Viewing Guardrail Details

Click any guardrail to see:

  • Runtime Constraint — The specific guidance given to your agent
  • Full Explanation — Why this principle matters
  • References — Academic and legal sources
  • Traditions — Cultural and philosophical origins

Changing Guardrails Later

You can update your agent's ethical guardrails anytime from the agent profile settings. Changes take effect immediately for future conversations.

Configuring Agency Oversight

After provisioning completes, you'll be presented with an agency selector screen. This lets you configure how much autonomy your agent has across twelve domains — from Strategy and Planning to Transactions and Scheduling.

Each domain supports three oversight levels:

  • Silent — Your agent acts freely without notifying you
  • Report — Your agent acts on its own but sends you a notification
  • Approval — Your agent proposes an action and waits for your explicit approval

All domains default to Report. Adjust them to match your comfort level — you can always change these later from your agent's settings.

For more details, see Understanding Agencies.

Execution Credentials

Some agents use coding CLIs or model providers to carry out work. You can manage those credentials after the agent is created.

  • CLI Authentication shows whether supported coding tools are installed, configured, and authorized
  • Inner Loop lets you choose the coding CLI and model the agent should use
  • Manage Credentials lets you add or update provider access
  • Require your own provider credentials prevents the agent from falling back to Kindship-managed credentials when you want execution tied to credentials you control

If a sibling agent already has a relevant credential preference, new-agent setup may inherit that expectation so the new agent starts with clearer execution requirements.

First Heartbeat

New agents get a first recurring heartbeat only when they are actually new. That first heartbeat gives the agent a starting rhythm without replaying first-run behavior for existing agents.

Use the Heartbeat tab after setup to see whether recurring work is landing on schedule.

Tips for Effective Agent Setup

Be Specific About Your Context

The more your agent knows about your situation, the better it can help:

  • Industry and market
  • Resources available
  • Past experience
  • Known constraints

Start with a Single Focus

While your agent can help with many things, start with one main goal. You can expand scope later as you get comfortable.

Use Natural Language

Talk to your agent like you would a colleague. No special commands or syntax required. If something isn't clear, your agent will ask for clarification.

Don't Worry About Perfection

Your Prime Directive and goals can evolve. It's better to start and refine than to spend too much time planning upfront.

Managing Multiple Agents

As your needs grow, you might want multiple agents:

  • Different projects — One agent per major initiative
  • Different roles — One for strategic planning, another for creative work
  • Team vs. personal — Separate work and personal agents

Each agent maintains its own context and memory, so conversations stay focused.

What Happens Next

After setting up your agent:

  1. Continue the conversation — Ask questions, give instructions, brainstorm ideas
  2. Build your planning structure — Add projects, processes, and tasks
  3. Return regularly — Your agent remembers everything and picks up where you left off
  4. Iterate and improve — Refine your approach based on what works

Troubleshooting

My agent doesn't understand my goals

Try being more specific. Give examples. Share context about your situation and constraints.

The responses are too long/short

Tell your agent directly: "Please give me shorter responses" or "I need more detail on this."

I want to change the Prime Directive

You can update it anytime. Just say "Let's update my Prime Directive" and discuss the changes.

My agent says it needs my own provider credentials

Open the agent workspace settings, review CLI Authentication, and add the required provider credential. If the agent is configured to require your own credentials, dispatch may stay blocked until that requirement is satisfied.

Next Steps